My newest attempts to create a "True Futa" Model, this time in Lora form and for various newer models (since the original was on the ancient SD 1.5), including Anima-Base, Krea2, & Illustrious
For when you just want a beautiful futanari/hermaphrodite/intersex individual without the "coinpurse" getting in the way of the best parts.
This model leverages the powerful existing NSFW knowledge of the new models and enhances it with additional training on a newly cultivated and tagged dataset of over 500+ images that I've been curating over the past several years.
The Anima version works well with many different Anima Checkpoints (not just Anime-Base) and can also be used alongside other Lora's, so its flexibility should let your creativity dictate the output!
The Krea 2 version is more experimental, as I have less familiarity with the model and just wanted to see what kinda of effect my dataset would have when training a lora on it (with some decent results)
The Illustrious version is using a slightly more refined Dataset from the previous Anima and Krea2 versions, with included regularization data to allow it to more easily differentiate between futanari and normal women (to help it overcome some of the limitations of the slightly older SDXL base model). I prefer to use it with the iLustMix 11.1 model, as it an extremely capable NSFW Illustrious based checkpoint.
Trigger Words are listed specifically for each model, they may not be required for the Lora to have an effect, but if the "concept" isn't expressing itself, including them is recommended
Description
Used the same dataset as the Anima version, recaptioned/retagged to better fit the requirements of Illustrious. All preview images generated on iLustMix 11.1
FAQ
Comments (5)
Finally! It was such a pain having to manually edit the picture and remove the balls to see her pussy. Sometimes they were gigantic just to make it more annoying. Anima works great.
My thoughts exactly! Tho ironically I kinda think my newer Illustrious version might be better than my Anima one, since I've been able to get a significantly greater diversity of poses and scenes out of it. (The ability to use Textual Embeddings on SDXL based models is also VERY nice, as there are a ton out there that play quite well with the True Futa Reborn lora in my testing)
I may consider training a new version of the Anima Lora with an even more refined dataset, but sorting and tagging takes a while (even with some level of LLM automation, since I usually still need to go back in and vet the results).
Looking at the sample images, the Anima and Illustrious versions have kind of unnatural placement of the penis, it looks more like a dildo glued to the top of the mons pubis, or even onto the lower stomach. The Krea2 version looks better with the underside of the penis at least being a part of the same structure as the clitoral hood, but it still seems to be located a bit too high. The erectile tissue of the penile shaft would more naturally be continuous down along the labia toward the perineum (if comparing to male anatomy).
Much of that seems to be due to the various base models (Anima & Illustrious) occasionally causing that differentiation (if you go through many of my sample outputs in the gallery you'll see that they DO manage to combine properly into a more gestalt whole in some images, but not consistantly).
Additionally, part of the "secret sauce" that appears to have made my earlier attempts possible is that I also included a small subset of custom generated and tagged training images of castrated femboys (about 30 or so), allowing the base concept of a "lack of testicles/scrotum" to be taken up more easily while training the Lora.
I've been working on refining my training dataset and have done several recent new internal test iterations of possible new versions for both the Anima and Illustrious versions, but have yet to get results that are better than I have already achieved (and in fact many of my attempts have ended up far worse, hence the lack of any additional releases).
If you have any suggestions, or have tagged dataset images you want to contribute to the training data, please do and I'll see what can be accomplished.
Oh yes, some of the samples do look better and the base model is probably the significant factor. Anima seems to struggle the most, while Krea2 is quite consistent.
I don't have anything better unfortunately, it seems to be a fairly rare concept and good quality depictions are very rare indeed. Your LoRAs are probably some of the best that we have so far and my critique was only a comparison to a hypothetical "ideal" of sorts.
There is a pair of older LoRAs for Pony and Illustrious here: https://civitai.red/models/1189148/true-futa-no-testicles and a few images in the gallery look "right" to just about the same degree as your samples but other than that I'm not sure if I've ever seen a really good representation of this, except possibly some occasional old drawings on booru sites.
Maybe one possibility would be to include some images of "hyper" clitorises which actually emerge from between the outer labia, unless it then messes up the futa penile anatomy: https://civitai.red/models/2013277/joystickmajora